About the Atlas

Before you build an onchain product, study what has already been tried.

What this is

The Onchain Experiment Atlas is a catalog and work-in-progress encyclopedia of onchain mechanism-design experiments. The catalog currently contains 450 candidate records. Its research drafts, diagrams, and simulations are inputs to an editorial review process; their presence in the repository does not mean that a human reviewer has approved them.

Methodology

• Primary sources first. Contracts and verified source code, official documentation, founder retrospectives, governance posts, audits and postmortems — in that order — before news coverage or commentary. A source list is not the same as claim-level citation. An entry cannot pass the publication gate until its material factual claims are linked to structured, verified source records.
• No fabrication. The editorial rule is to label missing information “Unknown / not found” rather than guess. Unreviewed generated drafts can still contain mistakes or unsupported statements, which is why draft status and second-person review are enforced separately.
• X-account verification. Every account carries a verification label: verified-by-official-website, verified-by-project-documentation, verified-by-founder- statement, strongly-inferred, unverified, or disputed. Unverified, inferred, malformed, or missing handles are not presented as verified links. Treat anything below “verified” with appropriate caution.
• No forced verdicts. Outcomes use a graded vocabulary (major success, partial success, technically-successful-commercially-unsuccessful, failed, exploited, abandoned, ongoing, too early to judge) rather than a binary.
• Research process — read this before citing us. Every entry was drafted by AI research agents working from primary sources under the rules above, then put through two further automated passes: a quality audit that graded completeness and sourcing, and an editorial pass that checked how wrongdoing is described. No human has reviewed all 450 entries line by line. Treat this as a well-sourced v1 reference and a starting point for your own research — not as a peer-reviewed authority. Errors are possible; corrections are prioritized.

Publication standard

Records move through draft → review → published, and the build fails closed: a record that requests publication without clearing the checks below does not ship.

Enforced on every published page: all seven required sections present, at least one real cited source, at least three reusable lessons, a launch date that is either precise or explicitly labeled approximate/unknown, an outcome drawn from the approved vocabulary, and valid identifiers, links, and social handles.

Not yet met corpus-wide, and deliberately disclosed rather than hidden: claim-by-claim citation records, complete source metadata (publisher, author, access and archive dates), structured on-chain contract evidence, and a named independent human reviewer for each entry. These are the standard this project is working toward, and pages will be upgraded as that research is completed. Until then, entries are labeled as AI-researched drafts rather than presented as reviewed scholarship.

Editorial disclaimers

• Every “Redesign” section is editorial hypothesis — an analytical thought experiment, not a statement of fact and not financial, legal, or investment advice.
• Simulations are illustrative draft models. Defaults may be synthetic, values and timing may be randomized, and they are never live onchain data. A simulation should not be used as evidence for a historical or financial claim.
• Editorial policy requires entries describing exploits, collapses, or alleged wrongdoing to distinguish allegations, charges, findings, convictions, and admissions. Until a draft has claim-level citations and reviewer approval, readers should not treat it as an adjudicated statement.
• Image and diagram rights must be recorded before publication. Current generated diagrams are intended as original explanatory work, not historical screenshots.

Corrections

If you are a founder, contributor, or researcher and something in your entry is wrong, flag it publicly on X and it will be reviewed against sources. Corrections that come with primary-source evidence are applied fastest.

Missing an experiment?

Search first — the Atlas covers 450 experiments from 2013 to 2026. If it’s genuinely missing, suggest it. And if the mechanism you’re looking for doesn’t exist yet, build it: apply to Alliance.